School Discipline Practices: Predictors of Juvenile and Adult Patterns of Offending Among At-Risk Children
Methods: The current study used an administrative data set containing educational, juvenile justice, and adult correctional information about 407,800 children enrolled in grades 7-12 in public schools in a state in the Deep South. The overall sample The study sample was drawn from a 10-year cohort (1996-2008) of children in grades 7-12 (N=45,799) who were identified as early starters (juvenile offense record; n=14,346;31.4%), late starters (adult offense record; n=17,105;37.3%), adolescence-limited (juvenile offense record, only; n=10,126;22.1%), and life-course-persistent (juvenile plus adult offense records; n=4,222;9.2%). Binary logistic regression (LR) was used to determine the combination of variables that best predicted four distinct patterns of offending: early starters, late starters, adolescence-limited, and life-course-persistent. Using the statistical software, STATA, four models were tested and predictors were entered in three blocks. Three demographic variables (gender, race, poverty status) were entered in the first block. The second block included the predictor, out-of-school expulsion. The last block included nine additional predictors: out-of-school suspension, in-school suspension, in-school expulsion, attendance, truancy risk, dropout status, school transitions, highest grade completed, and ELA standardized test scores.
Results: The overall sample of 407,800 children was composed of White (51.7%) and African American (44.1%) boys (52.2%) and girls (48.8%). LR results showed that Model 1 was statistically reliable in distinguishing between having and not having a juvenile offense record (-2 Log Likelihood =-50056.215; X2(14)=24109.71,p<.001). Results indicated that Model 2 was statistically reliable in distinguishing between having and not having an adult offense record (-2 Log Likelihood=-60677.242;X2(14)= 20634.12; p<.001). LR results showed that Model 3 was statistically reliable in distinguishing between having and not having only a juvenile offense record (-2 Log Likelihood=-40516.401; X2=13810.37;df=14;p<.001). Results indicated that Model 4 was statistically reliable in distinguishing between having both juvenile and adult records and not having both types of records (-2 Log Likelihood =-18223.99; X2=10526.95; df=14; p<.001). Wald statistics showed that in-school and out-of-school suspension and out-of-school expulsion significantly predicted each of the four patterns of offending. Odds ratios, however, for expulsion indicated moderate to large changes in the likelihood of having a juvenile record, an adult record, only a juvenile record, and both juvenile plus adult records; at Exp(B) = 11.2, 4.7, 8.0, and 7.8, respectively.
Conclusions and Implications: Results underscore the importance of implementing educational and psychosocial interventions to minimize expulsion from school. Additional longitudinal research is needed to understand how school discipline policies disproportionately affect children from minority groups.